# $29 or $30? What charm pricing really does to software sales

Source: https://leverprice.com/blog/charm-pricing-29-vs-30
Updated: 2026-11-17

> Nine-endings can lift demand: they did in mail-order catalog experiments and at Lyft. But the profit gains measured are a few percent at most, an online experiment and a replication found no effect, and one software test saw worse retention from 9-ending buyers. $30 beats $29 on revenue unless it loses over 3.3% of buyers, so test it.

$30 makes more money than $29 unless the extra dollar costs you more than 3.3% of your buyers. That's the whole decision, and the research on [charm pricing](https://leverprice.com/glossary/charm-pricing) is really about whether a 9 can move that many people. In some settings it clearly can. For software subscriptions we found no peer-reviewed test at all, and the effects measured elsewhere are smaller and messier than the single headline numbers that get passed around.

## Does charm pricing actually work?

Yes, in some settings, and the best evidence comes from field experiments where real customers saw different price endings.

The classic study is Eric Anderson and Duncan Simester's 2003 paper in Quantitative Marketing and Economics ([paper](https://doi.org/10.1023/A:1023581927405)). They ran three field experiments at two women's clothing catalogs, printing different prices for the same items in different versions of the catalog ([Simester's review](https://conference.nber.org/confer/2015/FEs15/Simester.pdf)). A $9 ending raised demand in all three experiments. The lift was bigger for items new to the catalog than for items it had sold in earlier years, and there was some evidence it shrank when the item also carried a "Sale" sign.

In the example the authors gave Harvard Business Review readers, a dress sold about a third more at $39 than at $34 ([HBR](https://hbr.org/2003/09/mind-your-pricing-cues)). A 1999 write-up of their price-ending research describes the same pattern at higher prices: demand was nearly identical at $44 and $54, and significantly higher at $49 ([Chicago Booth Review](https://www.chicagobooth.edu/review/yours-only-1999)). The authors read this as the 9 telling shoppers with little price knowledge that they were looking at a good deal. A prospect comparing three SaaS pricing pages side by side knows a lot more about prices than a catalog shopper looking at a new dress, which by the same logic should weaken the effect.

Manoj Thomas and Vicki Morwitz explained the mechanism in five experiments published in the Journal of Consumer Research in 2005 ([paper](https://doi.org/10.1086/429600)). People judged a 9-ending price as noticeably smaller than a price one cent higher only when the leftmost digit changed, as with $2.99 and $3.00, and the effect was stronger when the two prices being compared were close. $29 versus $30 changes the leftmost digit. $28 versus $29 doesn't, so the theory predicts nothing special there.

Recent evidence at scale comes from Lyft. John List, Ian Muir, Devin Pope and Gregory Sun analyzed more than 600 million ride sessions, then ran a two-month field experiment with 21 million riders, published in the Review of Economic Studies in 2023 ([paper](https://doi.org/10.1093/restud/rdad014)). About half of the drop in demand as prices rose happened in jumps at whole-dollar thresholds, such as a ride going from $13.99 to $14.00, and riders reacted to a one-cent cut below a dollar as if it were about 50 cents.

Retail data agree. Avner Strulov-Shlain studied scanner data on 3,500 products sold by 25 US chains and estimated that shoppers respond to a one-cent increase from a 99-ending price as if it were more than 20 cents (Review of Economic Studies, 2023, [paper](https://doi.org/10.1093/restud/rdac082)).

## How big is the charm pricing effect, really?

The jump in demand at a left digit can be large, but the extra money a business makes from it is small. The 99-cent change Lyft actually ran, which moved prices between $X.00 and $X.09 down to the 99-cent mark below, was worth about 0.39% in profit per user. The authors estimate that moving every price to the nearest 99-cent mark could add about 25 cents per ride, roughly $160 million a year at Lyft's 2019 volume, and only if the short-run effect lasts. Strulov-Shlain puts what the retailers in his data leave on the table by under-using 99-endings at 1% to 4% of gross profit.

| Study | Setting | What was compared | Result |
|---|---|---|---|
| Anderson and Simester, 2003 | Three field experiments, two women's clothing catalogs | $9 endings against nearby prices, such as $39 against $34 | Demand up in all three; strongest for new items, weaker with a "Sale" sign |
| Thomas and Morwitz, 2005 | Five lab experiments | Prices like $2.99 against $3.00 | A 9-ending feels smaller only when the leftmost digit changes |
| List, Muir, Pope and Sun, 2023 | Lyft, 21 million riders | Prices from $X.00 to $X.09 cut to the 99-cent mark below | Half the demand slope at dollar thresholds; about +0.39% profit per user |
| Strulov-Shlain, 2023 | 3,500 products, 25 US retail chains | 99-endings in scanner data | A 1-cent rise from .99 acts like 20+ cents; chains forgo 1% to 4% of gross profit |
| Fenneman and colleagues, 2022 | Online experiment, 266 people, 4,788 purchase decisions | Just-below prices against round prices | No significant difference |
| ProfitWell (now Paddle), 2020 | About 10,000 potential buyers of subscription software | 9-endings against 0-endings | 15 to 20% more conversions under $200 a month, about zero above $5,000; about 10% worse net retention |

Two rows in that table found nothing or found a cost. Fenneman and colleagues ran a preregistered online experiment in which 266 people made 4,788 paid, incentivized decisions about buying lottery tickets, and just-below prices sold no better than round ones: 47.1% against 45.9%, not a significant difference ([PLoS ONE, 2022](https://doi.org/10.1371/journal.pone.0270850)). Among their explanations, the effect may be smaller than the literature suggests, or may mostly appear when people compare options side by side.

The only software data we could find comes from ProfitWell, now part of Paddle, in a 2020 blog post ([Paddle](https://www.paddle.com/blog/end-prices-in-9s)). An experiment with just over 10,000 potential buyers found conversion 15% to 20% higher with 9-endings for products under $200 a month, fading to about zero above $5,000 a month. Customers who converted on a 0-ending price had about 10% better net retention. The post doesn't publish the method, the sample per price, confidence intervals or whether the lift is relative, and its author says the retention data was thin. Treat it as a hypothesis worth testing.

## When do round prices beat prices ending in 9?

When the price has to signal quality, and possibly when the purchase runs on feelings more than calculation. The quality evidence is stronger than the feelings evidence.

Mark Stiving's model and data (Management Science, 2000) found that firms using high prices to signal quality tend to set round prices, and that higher-quality products carry rounder prices, especially in categories where buyers can't judge quality before purchase ([paper](https://doi.org/10.1287/mnsc.46.12.1617.12078)). Software, which you can't fully judge until you've used it, sits closer to that end than groceries do.

Robert Schindler and Thomas Kibarian tested price ads directly (Journal of Advertising, 2001). A 99-ending such as $49.99 instead of $50.00 made viewers more likely to see the price as low and as a discount, and it hurt the quality image of ads from higher-quality retailers ([paper](https://doi.org/10.1080/00913367.2001.10673654)). A 9 says "deal". That helps a bargain and works against a premium product.

The feelings story is weaker. Monica Wadhwa and Kuangjie Zhang reported in five studies that rounded prices like $100.00 made people more likely to buy when the purchase was driven by feelings (a camera for a family vacation), and non-rounded prices like $98.76 when it was driven by reasoning (the same camera for a class project) ([Journal of Consumer Research, 2015](https://doi.org/10.1086/678484)). A preregistered replication of their fifth study with 588 participants found neither the interaction nor the proposed "feeling right" mechanism, and concluded the effect is smaller than reported or depends on factors nobody has identified ([Royal Society Open Science, 2018](https://doi.org/10.1098/rsos.171127)).

One more finding matters for plan grids. Kenneth Manning and David Sprott found that switching two prices from round ($30.00 and $40.00) to just-below ($29.99 and $39.99) shifted choice toward the cheaper option, with the size of the shift depending on price level and shopping goals ([Journal of Consumer Research, 2009](https://doi.org/10.1086/597215)). If that holds on a [good-better-best](https://leverprice.com/glossary/good-better-best-pricing) page, 9-endings on every tier could nudge people down to your cheapest plan.

## Is charm pricing different for B2B software?

Probably, though nobody has measured it well. In Duncan Simester's review of 61 marketing papers reporting field experiments from 1995 to 2014, only five involved firms selling to other businesses ([Simester's review](https://conference.nber.org/confer/2015/FEs15/Simester.pdf)). What follows is reasoning from the evidence above; none of it has been measured directly.

B2B buyers often approve a total, not a sticker price. Eight seats billed annually come to $2,784 a year at $29 and $2,880 at $30: the leftmost digit is a 2 either way. Ten seats billed monthly are $290 against $300, and there the left digit does flip. Look at the number your buyer actually signs off on before you decide which ending matters.

The signal changes too. A 9 reads as a discount, and in the catalog studies it mattered most when shoppers had little else to go on. A team choosing a tool to run its work on is often checking reliability more than looking for a bargain, which by the same logic should weaken the 9.

Budget rules can matter more than perception. Example: if a team lead can expense anything under $50 a month without approval, $49 and $50 differ for reasons that have nothing to do with how the digits look. ProfitWell's numbers point the same way: the lift it saw faded above $200 a month and was about zero above $5,000.

## Is $29 vs $30 even the right question?

Usually not. Where researchers measured profit, the ending was worth a few percent at most, while the gap between $29 and the price that earns the most per visitor can be far bigger.

| Instead of $29, charge | Price increase | Buyers you can lose before revenue drops |
|---|---|---|
| $30 | +3.4% | 3.3% |
| $35 | +20.7% | 17.1% |
| $39 | +34.5% | 25.6% |
| $49 | +69.0% | 40.8% |

*Figure: Illustration: as price rises, the share of visitors who buy falls steadily while revenue per visitor rises, peaks and then falls. Illustration: conversion falls as price rises, and revenue per visitor peaks in between. A $1 change moves you a tiny step along this curve.*

For $29 to beat $30, the round price has to lose more than 3.3% of the buyers $29 would get. For $29 to beat $39, the extra $10 has to lose a quarter of them, which is a much bigger ask. If you're unsure where you sit on the curve, test the price level before you test the ending, and run your own pairs through the [price increase calculator](https://leverprice.com/tools/price-increase-calculator).

## How do you test $29 vs $30 vs $35 on real visitors?

Show each new visitor one of the three prices, keep them on it, and compare revenue per visitor after refunds. Plan for $29 and $30 to stay too close to call, and make sure the test still teaches you something when they do.

1. Assign the price on your server, so the pricing page and the checkout always agree and bots don't pollute the numbers.
2. Keep each visitor on the price they saw first.
3. Measure [revenue per visitor](https://leverprice.com/glossary/revenue-per-visitor) after refunds, not conversion rate. $30 can convert slightly worse and still earn more.
4. Run whole weeks, so weekday and weekend traffic both count.
5. Before you lock in a winner, compare how each price's buyers retain over two or three billing cycles, since ProfitWell's warning was about retention.

The hard part is sample size. At a 2% conversion rate, detecting the 3.3% gap between $29 and $30 with 95% confidence and 80% power takes about 695,000 visitors per price; at 5% it's about 270,000. The 17.1% gap that decides $29 against $35 takes about 24,000 visitors per price at 2% ([sample size calculator](https://leverprice.com/tools/price-test-sample-size-calculator)).

*Figure: Chart: visitors needed per price fall steeply as the difference you want to detect grows, for 2% and 5% conversion rates. Visitors needed per price at 95% confidence and 80% power. The 3.3% gap between $29 and $30 is off the left edge of this chart.*

[Leverprice](https://leverprice.com/price-testing) runs this as a bandit rather than a fixed three-way split. It shifts traffic toward the price earning the most per visitor as evidence comes in, and calls the test when one price is at least 95% likely to be best or when the remaining prices are practically equivalent, meaning the expected loss from picking either is under 0.5%. For $29 against $30 that second outcome is the likely one, and it's a useful answer: keep either and spend your traffic on whether $35 beats them both. [Bandits versus A/B tests](https://leverprice.com/blog/bandit-vs-ab-test-pricing) explains the trade-off.

Until you have your own data, this table is the decision rule:

| Your situation | What to do | Why |
|---|---|---|
| Self-serve product under $200 a month | Test the 9-ending against the round price, and compare retention by price for two or three billing cycles | ProfitWell's conversion lift showed up below $200 a month, with worse retention |
| B2B, priced per seat or paid on invoice | Use round prices unless a test says otherwise | Buyers approve totals, and the B2B evidence is thin |
| Premium product, or quality that's hard to judge before buying | Use round prices | Round prices go with quality signals (Stiving; Schindler and Kibarian) |
| Choosing between $29 and $30 at all | Add a bigger step such as $35 or $39 to the same test | Measured profit gains from endings are a few percent; the price level can be worth far more |

## Frequently asked questions

### Should a SaaS price end in .99?

We found no published evidence that $29.99 sells more software than $29 or $30. In Thomas and Morwitz's experiments the left-digit effect needed the leftmost digit to change, which happens between $29.99 and $30 but not between $29 and $29.99, and whole-dollar prices are easier to read on a plan grid.

### Does charm pricing work for annual prices like $299 vs $300?

The mechanism applies, because the leftmost digit changes. A yearly payment is a bigger, more deliberate purchase, the kind where ProfitWell saw the lift shrink, so test the yearly price on its own; [annual vs monthly discount](https://leverprice.com/blog/annual-vs-monthly-discount) covers how.

### Do 9-ending prices attract worse customers?

Possibly. ProfitWell found customers who converted on a 0-ending price had about 10% better net retention, and Schindler and Kibarian found 99-endings make a price read as a discount. Compare retention by price before you commit.

### How many visitors do I need to test $29 against $30?

About 695,000 visitors per price at a 2% conversion rate, or about 270,000 at 5%, to detect a 3.3% difference with 95% confidence and 80% power. Most products should test a bigger gap instead.
